Peer Review with AI & Author Stress Test: A Two-Way Protocol for Research Quality Pre-Verification

Title: Peer Review with AI & Author Stress Test: A Two-Way Protocol for Research Quality Pre-Verification This white paper introduces DAUL, a novel two-way protocol designed to leverage Large Language Models (LLMs) for adversarial peer review and author self-testing. Developed in early 2026, the study addresses the lack of transparency in AI-assisted reviews and the absence of robust self-verification tools for researchers. The protocol follows an "Attack-Defense" framework where an AI acts as a professional adversary, attempting to dismantle the paper's logic while the author defends it by strictly referencing the text. The study documents four detailed case studies involving Claude Sonnet 4.5 and Grok 4.1 Thinking, testing across two distinct domains: foundational physics and meta-science. Key findings include:- A 100% consistent pattern of AI behavior: initial skimming and misapplication of criteria followed by significant adjustment and deep understanding once challenged with specific evidence.- Demonstrated capabilities of LLMs in "deep reasoning" within non-English (Vietnamese) academic contexts.- A "Clean Slate" protocol to ensure interaction independence and minimize bias. This dataset includes the full white paper and links to verifiable transcripts and video recordings, providing a transparent benchmark for AI academic reasoning. DATASET: TRANSCRIP: 4 PDF FILE BÉO. (2026). Adversarial AI Peer Review Transcripts: Attack-Defense Protocol Testing. Zenodo. https://doi.org/10.5281/zenodo.18211530 VIDEO: https://www.youtube.com/playlist?list=PLJ3dkTh-U9b8OLdws0M6N_5bjGNsaAGzf LISENCE DATASET: Dual LISENCE LISENCE THIS PAER: CC BY 4.0 CONTACT: BÉO - INPENDENT RESEARCH beo@beolabs.org